Adsorption-energy prediction on OC20 16k Dense
-7.38Improvement MAE vs CatBERTa (%)GAP-CatBERTa
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GAP-CatBERTaPretrain Data=OC20 (460k), Fine-tuning Data=OC20-Dense (16k), Inference Input=text-only2025.12 | -7.38 | — | — | 7.3 | |
| QE-Catalytic*Pretrain Data=OC20 (460k), Fine-tuning Data=OC20-Dense (16k), Inference Input=text-only2025.12 | -12 | — | — | 11.2 | |
| QE-CatalyticPretrain Data=OC20 (460k), Fine-tuning Data=OC20-Dense (16k), Inference Input=3D+text2025.12 | -21.2 | — | — | 14.9 |